In many analogue circuit arrangements there is a requirement for accurate and constant reference voltages and currents. In the case of reference currents, these are often generated using a reference voltage source and a resistor. In integrated analogue circuits, chip space is at a premium and it is common to integrate only one, or a small number of such source/resistor combinations into a chip and to generate other reference currents from the basic reference current(s) using current mirrors. Current mirrors have a relatively simple construction and do not occupy a large chip area. A typical current mirror, comprising a pair of matched MOSFET transistors (a reference transistor T and a mirroring transistor M), is illustrated in FIG. 1 where t.sub.n is the reference current and lout is the mirrored output current. The current mirror may be extended by adding one or more additional current mirroring transistors in parallel.
One application for current mirrors is in digital to analogue (D/A) converters, where inaccuracies in the reference currents and voltages tend to dominate as a source of conversion errors over factors such as switch resistances and unmatched loads. A simple 4-bit D/A converter is shown in FIG. 2, and comprises a reference MOSFET transistor T and four mirroring MOSFET transistors M.sub.0 to M.sub.3 arranged in a current mirroring configuration. Transistors T and M.sub.0 have a channel width to length ratio of W/L whilst transistors M.sub.1, M.sub.2, and M.sub.3 have ratios of 2W/L, 4W/L, and 8W/L respectively. A switch (S.sub.0 to S.sub.3) is arranged in series with the drain of each MOSFET (M.sub.0 to M.sub.3), the switches being controlled by respective bits (a.sub.0 to a.sub.3) of the digital (binary) signal to be converted. a.sub.0 corresponds to the least significant bit (LSB) of the signal whilst a.sub.3 corresponds to the most significant bit (MSB). The switches (S.sub.0 to S.sub.3) are coupled together at one end and draw an output current I.sub.out depending upon the activation state of the switches.
Current mirrors such as that employed in the D/A converter of FIG. 2 copy or scale reference currents with varying degrees of accuracy depending upon the precise construction of the mirrors. In practice, errors will arise because the transistors of the current mirror are not identical (or are not scaled identically). The extent of this mismatch may be minimised by using the common-centroid method to determine the layout of the transistors. For example, for the 4-bit converter shown in FIG. 2, transistors T and M.sub.0 are provided by respective single transistor elements, whilst transistors M.sub.1, M.sub.2, and M.sub.3 are provided respectively by two, four, and eight parallel connected transistor elements (each with a channel width to length ratio of W/L). These elements are arranged in a geometric pattern as illustrated in FIG. 3 where transistor elements are shown by element number (and T represents the diode connected reference transistor). However, particularly for high bit D/A converters (e.g. 12 or 14 bits), the common centroid method does not ensure high accuracy for conversion of the MSBs. This may be due to the large separation between transistor elements making up the transistors corresponding to these bits. Whilst techniques such as laser trimming can be used to physically adjust the dimensions and properties of certain circuit elements, these are generally time consuming and expensive to implement.
There are many other applications where inaccuracies in the properties of field effect transistors can be a source of serious error. One example is the so called `pipeline` analogue to digital (A/D) converters in which field effect transistors are incorporated into subtractor, multiplier, and comparator circuits. The precision of such a converter is critically dependent upon the 30 operating precision of the transistors in these circuits.
In high quality applications D/A and A/D converters must be capable of operating at the Nyquist frequency. In the case of mobile communication devices, at present and near future data processing rates, use of converters in the intermediate-frequency (IF) band means an operating frequency of 10 MHz or greater. Whilst converters based upon the architecture of FIG. 2 can operate at such high frequencies, the conversion accuracy is often unsatisfactory for high bit converters. On the other hand, converters based on alternative architectures, such as the sigma-delta (or oversampling) converter, offer high accuracy and high bit conversion but suffer from slow speed (operating at only a few MHz). With these or other converters, the large chip area occupied by the converter as well as high power consumption can also become a problem where large scale integration is desired.